Preference
PRISM Alignment
Human ratings and selected-response flags for 21 generating models in PRISM conversations.
53,926items
21subjects
Human-written text: CC-BY-4.0. Model outputs: CC-BY-NC-4.0 plus provider terms. Retain source attribution, data clause and no-re-identification requirements in raw/README.md.license
preferencedomain
textmodality
item-level responses released
Response matrix
Loading response matrix…
lowhighUnobserved
Scale: Item-specific grading scales